HA with JDE / Oracle 8.1.7/ Network App

[Donna Yedziniak]

We are running HP-UX 11 on two N series servers and are looking at our HA options for our environment. Hopefully someone else out there has a setup like this. We are running JD Edwards on Oracle 8.1.7, Enterprise Edition. Our storage will be clustered Network Appliance Netfiler 840's. (Which don't support RAW storage). We are currently migrating off a mainframe (9 TB) and the database size will be (estimated) 2 -3 TB.

What are your experiences out there?
Do we cluster the two N's with MC/ServiceGuard?
What would be good for Oracle?
We are an NT / Mainframe/ AS400 shop and the HP-UX world is new to us and we are not sure which direction we should head.
Our experience on Oracle is on the NT platform with small databases up to 20 GB.
Thanks,
Donna

[Printaporn_1]

MC/service guard with Oracle toolkit is the most suitable

[Donna Yedziniak]

Thanks for you reply. Is the product called "oracle toolkit" and is this part of MC/Serviceguard or a separate HP product?
What type of replication or standby database does it create?

Thanks,
Donna

[Volker Borowski · Accepted solution]

Hi Donna,

ServiceGuard is HP-product for clustering and has no idea of oracle. You define packages of processes and filesystems on a virtual IP-host that is "switching" to another server, if your main box is down.

This implies a Restart of the database.

I would check the need for your application. A cluster is a fast solution to be online again soon.

BUT, if you talk about TB, you should consider application faults as well.

DROP TABLE OPEN_INVOICES;

...will be a fine SQL-statement, syntactic correct, consistent and quickly executed and instantly clustered on all physicly intact mirror disks :)

This user fault will need a restore of the entire DB, since you'll need to do a point in time recovery. I'll bet for TB this will take some time....

For a DB of this size, I would consider a shadow DB, which is kept quite a while (not minutes) behind the leading DB. I would say at least 8 hours, may be even for two fixpoints a day (08:00 and 19:00 or so).

If your primary DB crashes physically, you'll need to do a forward recovery, which will need some time (some 20 minutes, may be up to a few hours). If you have got a logical fault like the above, you can open the shadow instantly at one of your fixpoints instead of restoring your TBs.

If you want to be safe for both, you'll need both (cluster/mirror AND shadow). One is no replacement for the other, but if "short" downtime is acceptable, a shadow is the better solution, because you have a chance to catch user faults as well, which no mirror/cluster will ever do.

Shadowing can be done with Oracle-Tools or 3rd party tools (libelle or something like this).

Something to think about
Volker